At NAB Show New York, a session on the evolution of AI in radio and podcasting confronted how automation is reshaping production, personalization, and brand safety. With roughly 11,500 in attendance, panelists emphasized disciplined experimentation over hype, arguing that trust, disclosure, and human oversight must anchor any rollout.
Bonneville International’s Sheryl Worsley outlined practical wins from AI transcription, translation, and AI-read articles that accelerate publishing without replacing staff. She stressed strict data governance: Bonneville’s tools run in a closed environment to prevent leakage and protect intellectual property, a nonnegotiable for broadcasters handling proprietary audio and audience data.
SoundStack’s Matt Kellogg detailed brand safety advances through a partnership with Barometer that grades podcast content risk to help advertisers avoid adjacency landmines. He warned that blunt models can overcorrect—particularly in true crime and news—unless human reviewers calibrate edge cases and protect legitimate inventory. The group agreed that machine triage plus human context yields the most defensible standards.
Ordo Digital’s Jon Accarrino urged teams to separate utility from novelty by setting success metrics (latency, error rate, lift) before deployment. He recommended lightweight pilots, red-team testing for prompt injection and data exfiltration, and clear rollback plans.
Panelists flagged the rise of “AI slop,” a glut of low-quality synthetic output that undermines audience trust and advertiser confidence. Disclosure was a recurring theme: when using synthetic narration, label it plainly and preserve a consistent brand voice. The discussion also touched on cloned and synthetic voices entering broadcast workflows; meteorologist Amy Freeze was cited as an early adopter who built an AI version of herself to deliver personalized forecasts from National Weather Service data—an example of augmentation rather than replacement.
Takeaways for stations and networks: lock down data paths; require human sign-off on sensitive content; publish an on-air and digital disclosure standard; score creative with preflight checklists for accuracy, rights, and safety; and track downstream results so AI features prove their value. For marketers, align targeting and brand safety settings to category norms, monitor false positives, and insist on transparency from vendors.
The consensus: try it, measure it, and keep people in the loop—because credibility is the moat AI cannot build on its own.